A novel gradient-based matching via voting technique for vision-based structural displacement measurement

The effectiveness of vision-sensing system for field applications in remote measurements of structural displacement can be adversely affected by environmental factors, leading to an incomplete recording of a moving feature target in recorded video frames. The result is a non-ideal solution using traditional template matching techniques, which presents significant challenges for accurate measurements. This paper proposes a novel gradient-based matching via voting (GMV) technique to overcome environmental and operational conditions and provide reliable tracking of moved targets with different degrees of feature loss. When evaluating the region similarity between the template and a subset (called the overlapping region) of video frames in which the moved target is located to obtain a similarity score of the subset, the traditional template matching does not ascribe weights to the similarity at the pixel level, whereas GMV uses a voting scheme to weight all the pixel similarities. Three experiments were used to test GMV, evaluating its tracking accuracy and verifying its robustness with different levels of target feature loss based on a comparison with two commonly used template matching techniques. GMV proved to be capable of retrieving accurate displacement data with a high degree of accuracy, even for a moved target with up to 90% feature losses.